The transcription factor p53 is a central tumor suppressor protein that controls DNA repair, cell cycle arrest, and apoptosis. About half of human cancers have p53 mutations, and restoring p53 function in advanced tumors leads to tumor regression. Significantly, the large majority of these tumors produce full-length p53 proteins that have lost their tumor suppressor function due to single amino acid changes. Therefore, an attractive new approach to systemic cancer therapy is pharmacological reactivation of p53 cancer mutants. Reactivation of p53 cancer mutants is feasible because we and others have shown that introducing additional mutations (second-site suppressor mutations) can restore activity to the otherwise inactive p53 cancer mutants. In addition a few promising small molecule drug leads with unknown mechanisms of action have been reported to reactivate p53 cancer mutants. The challenge is to understand structural changes that lead to reactivation of p53 cancer mutants and to induce such changes through small molecules. This is a complex problem due to the diversity of clinically relevant single amino acid changes found in p53 cancer mutants. We propose computational approaches based on machine learning that are supported by high-throughput biological strategies. We use novel saturation mutagenesis approaches to catalogue p53 rescue mutations for the 50 most relevant p53 mutants found in human cancer. The genetic data are used in an active learning scheme to train a computational classifier, that is based on modeled atom-level structural features, to predict which structural changes lead to reactivation of any given p53 cancer mutant. We further propose to apply this improved computational classifier to discover small molecules that induce similar structural changes and test these pre-selected compounds in a biological assay for p53 cancer mutant reactivation. Data obtained from these in vivo experiments will be used to further improve the computational predictions for small molecules. In summary, we use genetic functional data to train a structure-based classifier to predict p53 activity based on an internal representation of modeled structural changes. The classifier will then be used to predict reactivation of p53 cancer mutants by small molecules with the aim to identify cancer drug leads. The proposed research has high impact on biomedical research and public health. About 250,000 US deaths yearly are due to tumors with full length but mutated and inactive p53. The long-term goal of this research, a drug that reactivates mutant p53, could prevent or delay these deaths. PUBLIC HEALTH RELEVANCE: The tumor suppressor protein p53 is the single most important protein to prevent cancer, and half of all human tumors produce a defective p53 protein. Pharmacological reactivation of the defective p53 proteins could potentially prevent or delay over 250,000 cancer deaths yearly in the US alone. This proposal uses genetic strategies and computational approaches to predict p53 reactivation by small molecules that can eventually develop into cancer drugs.